Deployment Strategies

A deployment strategy is really a question of how much information you buy before you're fully committed, and how expensive backing out is if that information is bad. Rolling deploys buy almost nothing (all pods eventually run the new version, fast) but are cheap; canaries buy a lot (real traffic against a small blast radius) at the cost of complexity; blue-green buys instant rollback at the cost of running double infrastructure. Pick based on the blast radius of being wrong, not on what's trendy.

The deploy itself is rarely the risky part — not having a tested way back out is.

Every deployment strategy is a trade between how much you learn before full exposure and how fast and cheaply you can undo it if you learn something bad.

1. Match the strategy to the blast radius, not the org's default